Wrong GPU announced in my computer specs
21 Feb 2025 12:15:46 UTC
Topic 232127
(moderation:
Hallo!
In my computer specs, the noun for E@H used GPU on my CPU is specified. For crunching E@H I’m using the much more capable separate GPU RX66000 only, by a small file in my E@H setup. In my Event Log my GPUs are listed as:
21/02/2025 10:19:44 | | OpenCL: AMD/ATI GPU 0: AMD Radeon(TM) Graphics (driver version 3628.0 (PAL,LC), device version OpenCL 2.0 AMD-APP (3628.0), 12477MB, 12477MB available, 563 GFLOPS peak)
21/02/2025 10:19:44 | | OpenCL: AMD/ATI GPU 1: AMD Radeon RX 6600 (driver version 3628.0 (PAL,LC), device version OpenCL 2.0 AMD-APP (3628.0), 8176MB, 8176MB available, 7326 GFLOPS peak)
I assume GPU 0 was chosen for the computer specs because of its 50% bigger memory than for GPU 1. But the specified crunching capability of GPU 0 is just 7.3% of that from GPU 1. So GPU 1 is the much more capable GPU. Why isn’t this chosen in my computer specs?
I believe this wrongly chosen GPU 0 is the reason why I get so little spare tasks in my buffer.
How can this become corrected?
Kind Regards and happy crunching
Martin
P.S. : By the way. From evaluating of the last 408 tasks of O3AS_BU, I calculated a crunching capability for this RX6600 of (7783,1 +/- 8,2) [GFLOPS], assuming, that the CPU is crunching almost 100% parallel to the GPU. If they are alternating, the capability would become about 32% higher, as the ratio of (CPU/GPU)Time is (24,2122 ± 0,0050) [%]. GPU load is about 95% with 4 tasks parallel. These precise values become evaluated by fitting Gaussian curves into the probability density distributions.
the priority hierarchy for
)
the priority hierarchy for detecting the "best" GPU for AMD in BOINC is as such:
// criteria:
//
// - double precision support
// - local RAM
// - speed
//
https://github.com/BOINC/boinc/blob/master/client/gpu_amd.cpp
speed is the last priority. so it only comes into play if both GPUs have exactly the same VRAM amount. if you are able to reduce the VRAM allocated to your integrated iGPU, that would probably solve that problem. or if you're not using the iGPU at all, you can disable it in the BIOS.
_________________________________________________________________________
Hello Ian&Steve ! Thanks
)
Hello Ian&Steve !
Thanks for your quick answer and explanation.
I’ll try to reduce the VRAM for the onboard GPU in the BIOS. But I can’t do it immediately.
Kind regards and happy crunching
Martin
Martin, You could always
)
Martin,
You could always edit your cc_config file and have it ignore the GPU 0 device if you just want to try running the RX6600 alone. Something along the lines of -
<cc_config>
<options>
<use_all_gpus>1</use_all_gpus>
<ignore_ati_dev>0</ignore_ati_dev>
</options>
<log_flags>
<task>1</task>
<file_xfer>1</file_xfer>
<sched_ops>1</sched_ops>
</log_flags>
</cc_config>
------------
'use_all_gups - ' not needed I also used a <ncpus>8 line in mine but unless you need that ..... search for boinc_config gives you all the options.
On some of my machines the RX amdgpus are detected as GPU device 0 so I have the opposite of above. My cpu - gpu is detected as GPU 1 so I ignore that number. you can do the same with intel or nvidia specified as well.
Maybe you want to run both though and your other idea will work - ?
-Mike
Hello Mike!Thanks for
)
Hello Mike!
Thanks for your hint. I’ll add the last part to my config file.
Kind regards and happy crunching
Martin